restat-tables-figures
GitHub专为REStat论文设计出版级表格与图表，确保主结果清晰自包含。遵循AEA风格，强调标准误、置信区间及独立注释，指导DID/RD等场景的可视化呈现，不执行底层估计。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill restat-tables-figures -g -y
SKILL.md
Frontmatter
{
"name": "restat-tables-figures",
"description": "Use when building or revising the exhibits of a The Review of Economics and Statistics (REStat) manuscript so the main result is legible in one table or figure in AEA\/REStat house style. Designs publication-grade exhibits; it does not run the underlying estimation."
}
Tables & Figures (restat-tables-figures)
When to trigger
- The main result is buried across several dense tables instead of one clear exhibit
- Tables lack self-contained notes (a reader must hunt the text for the spec)
- Figures are decorative rather than load-bearing, or omit uncertainty
- You are preparing exhibits for the online appendix vs the main paper
The REStat exhibit bar
A REStat reader should grasp the headline estimate from one main exhibit — a table or, increasingly, an event-study / RD / coefficient-stability figure with a confidence band. REStat follows standard economics house style: significance stars are permitted, but standard errors are expected in parentheses and the exhibit must stand on its own. Every table and figure carries a self-contained note stating the sample, the specification, the unit of observation, the inference (clustering level), and what the reader should take away. The main paper is self-contained; supporting detail goes to the online appendix (≤20 pages; source map refreshed 2026-06-20).
Exhibit design rules
| Element | REStat-compliant practice |
|---|---|
| Headline result | One main table or one figure a reader can read in 30 seconds |
| Standard errors | In parentheses below coefficients; clustering stated in the note |
| Significance | Stars allowed, but never instead of SEs; report the SE always |
| Notes | Self-contained: sample, spec, unit, N, inference, source |
| Figures | Load-bearing (event study with leads, RD plot, coefficient stability) with CIs/bands |
| Decimals | Consistent, sensible precision; no spurious digits |
| Appendix | Robustness and construction detail → online appendix (≤20 pp); main paper self-contained |
| Reproducibility | Every number traces to a script in the Harvard Dataverse package (restat-replication-package) |
Which exhibit for which design
- DID / event study: an event-study plot with leads and lags and a confidence band is often the single most persuasive exhibit — flat pre-trends visible at a glance.
- RD: a binned scatter with the fitted polynomials and the discontinuity, plus the estimate table with bandwidth/robust CIs.
- IV / shift-share: a first-stage table or figure alongside the structural estimate; do not hide a weak first stage.
- Measurement papers: a validation figure (new measure vs benchmark) earns its place as a main exhibit.
Execution bridge (StatsPAI / Stata MCP)
Generate exhibits from the fitted result, not by retyping numbers. Full map:
execution-with-mcp. REStat is applied econometrics/empirical micro — the home of careful identification; DiD/IV/RDD with weak-IV-robust CIs.
- Tables:
etable(multi-model) ordid_summary_to_latexstraight from theresult_id. - Figures:
plot_from_result/enhanced_event_study_plot/event_study_table— axis units and the SE/clustering note baked in. - Every note names the estimator + clustering and states the magnitude in interpretable units.
See a full fitted-result → exhibit chain in the JF execution walkthrough.
Checklist
- The headline estimate is legible from ONE main exhibit
- Standard errors in parentheses; clustering level stated in the note
- Stars (if used) accompany, never replace, standard errors
- Every table/figure note is self-contained (sample, spec, unit, N, inference, source)
- Figures are load-bearing and carry uncertainty (CIs / bands)
- Main paper self-contained; supporting exhibits moved to online appendix (≤20 pp)
- Every reported number is reproducible from the deposited code
Anti-patterns
- A wall of coefficients with no single readable headline exhibit
- Stars with no standard errors (REStat expects the SE shown)
- Table notes that force the reader back into the text to learn the specification
- Decorative figures without confidence bands or a clear takeaway
- Overstuffed main paper that should have pushed detail to the online appendix
- Exhibit numbers that do not match what the deposited script regenerates
Worked vignette: one figure carrying the paper (illustrative)
A staggered-DID paper had its main result spread over three tables: a TWFE column, a Callaway–Sant'Anna column, and a robustness table. A REStat editor wanted to see the identification at a glance. The fix collapsed the headline into a single event-study figure — event-time on the x-axis, the estimated effect with a 95% confidence band on the y-axis, leads visibly flat and centered on zero, lags showing the dynamic effect. The accompanying table kept only the pooled post estimate with its SE and clustering note. A reader now sees parallel-trends support and the effect in one exhibit; the three old tables moved to the online appendix. Self-contained note states sample, estimator, unit, N, and clustering — nothing to hunt for in the text.
Output format
【Headline exhibit】Table/Figure [n]: [what it shows in one line]
【SE + inference】SEs in parentheses; clustering at [level]; stars: [used/not]
【Figure load-bearing?】[event-study / RD plot / validation / coef-stability] with CIs? [Y/N]
【Notes self-contained?】sample/spec/unit/N/inference/source present? [Y/N]
【Main vs appendix】main exhibits: [...]; online appendix (≤20pp): [...]
【Reproducible?】each number ← script in Dataverse package? [Y/N]
【Next step】restat-writing-style
Version History
- 1839142 Current 2026-07-05 14:22
